[jax-rs-spec users] Re: Non blocking IO: Publishers and Subscribers

From: Sergey Beryozkin <>
Date: Thu, 23 Mar 2017 11:42:46 +0000

Hi Pavel

Thanks for the update, initial question is, what does it mean having a
subscriber is not enough as far as the JAX-RS resource method processing
for ex a POST request the NIO way, can you provide a code example ? I
think I might understand, the resource will start pulling somehow, but
would like to clarify
Thanks, Sergey

On 23/03/17 08:47, Pavel Bucek wrote:
> Dear EG members,
> we are still making progress on nio proposal (please see "nio" branch
> on github [1]). Feel free to send any feedback you might have, or
> propose any changes.
> (nio branch has Publisher renamed to Source, but I'll use term
> Publisher in this email to avoid confusion).
> When we evaluated how other frameworks are using similar API, we
> noticed that "public" APIs are usually Publishers, not subscribers.
> Thus we are trying to remove all exposed subscribers from Nio proposal
> and replace it by providing Publishers or returning Publishers from
> resource methods / nio body workers.
> This has several consequences, derived from the stream nature of data
> we are usually processing (request/response entities). For example, we
> have to be sure that subscriber for any entity will get whole entity
> (not a single byte can be skipped). That brings us to couple of rules
> for framework provided entity publishers:
> a) Publisher won't start publishing unless there is a demand. Demand
> means subsciption.request(x) where x > 0. That means the sole presence
> of a subscriber/subscription is not enough.
> Reasoning: we want to avoid filling Publisher buffers or require ones.
> b) there will be a subscriber limit - only single subscriber is allowed.
> Reasoning: subscriber has to be certain that all bytes are consumed
> (when you'd skip initial part of the message, you can throw that out.
> Subscribing 2 subscribers with a guarantee that it will read complete
> entity is "hard" (puts caching requirements on the publisher, which
> might not be simple to fulfill). Also, when a user want's to have
> another subscriber for logging, it would need to be a processor and do
> a "chain" instead of a "tree". Another reason is that using one
> ByteBuffer instance among multiple readers is not possible - we'd need
> to do some kind of view and manage references.
> Another rule might need to apply from application returned publishers:
> When the publisher is created by the resource method, not based on the
> request entity (i.e. handling HTTP GET method), it needs to be
> "caching" or "lazy", something like [2] or [3]. I'm still not sure
> whether 3rd party frameworks do commonly support Publishers similar to
> what was described as a) and b) - if there is anyone who is more
> familiar with flows, please let us know.
> Any comments or suggestions are welcomed.
> Thanks and regards,
> Pavel
> [1]
> [2]
> [3]